[POWERPC] ibmebus: Add device creation and bus probing based on of_device
[linux-2.6] / arch / powerpc / kernel / process.c
1 /*
2  *  Derived from "arch/i386/kernel/process.c"
3  *    Copyright (C) 1995  Linus Torvalds
4  *
5  *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
6  *  Paul Mackerras (paulus@cs.anu.edu.au)
7  *
8  *  PowerPC version
9  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
10  *
11  *  This program is free software; you can redistribute it and/or
12  *  modify it under the terms of the GNU General Public License
13  *  as published by the Free Software Foundation; either version
14  *  2 of the License, or (at your option) any later version.
15  */
16
17 #include <linux/errno.h>
18 #include <linux/sched.h>
19 #include <linux/kernel.h>
20 #include <linux/mm.h>
21 #include <linux/smp.h>
22 #include <linux/stddef.h>
23 #include <linux/unistd.h>
24 #include <linux/ptrace.h>
25 #include <linux/slab.h>
26 #include <linux/user.h>
27 #include <linux/elf.h>
28 #include <linux/init.h>
29 #include <linux/prctl.h>
30 #include <linux/init_task.h>
31 #include <linux/module.h>
32 #include <linux/kallsyms.h>
33 #include <linux/mqueue.h>
34 #include <linux/hardirq.h>
35 #include <linux/utsname.h>
36
37 #include <asm/pgtable.h>
38 #include <asm/uaccess.h>
39 #include <asm/system.h>
40 #include <asm/io.h>
41 #include <asm/processor.h>
42 #include <asm/mmu.h>
43 #include <asm/prom.h>
44 #include <asm/machdep.h>
45 #include <asm/time.h>
46 #include <asm/syscalls.h>
47 #ifdef CONFIG_PPC64
48 #include <asm/firmware.h>
49 #endif
50
51 extern unsigned long _get_SP(void);
52
53 #ifndef CONFIG_SMP
54 struct task_struct *last_task_used_math = NULL;
55 struct task_struct *last_task_used_altivec = NULL;
56 struct task_struct *last_task_used_spe = NULL;
57 #endif
58
59 /*
60  * Make sure the floating-point register state in the
61  * the thread_struct is up to date for task tsk.
62  */
63 void flush_fp_to_thread(struct task_struct *tsk)
64 {
65         if (tsk->thread.regs) {
66                 /*
67                  * We need to disable preemption here because if we didn't,
68                  * another process could get scheduled after the regs->msr
69                  * test but before we have finished saving the FP registers
70                  * to the thread_struct.  That process could take over the
71                  * FPU, and then when we get scheduled again we would store
72                  * bogus values for the remaining FP registers.
73                  */
74                 preempt_disable();
75                 if (tsk->thread.regs->msr & MSR_FP) {
76 #ifdef CONFIG_SMP
77                         /*
78                          * This should only ever be called for current or
79                          * for a stopped child process.  Since we save away
80                          * the FP register state on context switch on SMP,
81                          * there is something wrong if a stopped child appears
82                          * to still have its FP state in the CPU registers.
83                          */
84                         BUG_ON(tsk != current);
85 #endif
86                         giveup_fpu(tsk);
87                 }
88                 preempt_enable();
89         }
90 }
91
92 void enable_kernel_fp(void)
93 {
94         WARN_ON(preemptible());
95
96 #ifdef CONFIG_SMP
97         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
98                 giveup_fpu(current);
99         else
100                 giveup_fpu(NULL);       /* just enables FP for kernel */
101 #else
102         giveup_fpu(last_task_used_math);
103 #endif /* CONFIG_SMP */
104 }
105 EXPORT_SYMBOL(enable_kernel_fp);
106
107 int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs)
108 {
109         if (!tsk->thread.regs)
110                 return 0;
111         flush_fp_to_thread(current);
112
113         memcpy(fpregs, &tsk->thread.fpr[0], sizeof(*fpregs));
114
115         return 1;
116 }
117
118 #ifdef CONFIG_ALTIVEC
119 void enable_kernel_altivec(void)
120 {
121         WARN_ON(preemptible());
122
123 #ifdef CONFIG_SMP
124         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
125                 giveup_altivec(current);
126         else
127                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
128 #else
129         giveup_altivec(last_task_used_altivec);
130 #endif /* CONFIG_SMP */
131 }
132 EXPORT_SYMBOL(enable_kernel_altivec);
133
134 /*
135  * Make sure the VMX/Altivec register state in the
136  * the thread_struct is up to date for task tsk.
137  */
138 void flush_altivec_to_thread(struct task_struct *tsk)
139 {
140         if (tsk->thread.regs) {
141                 preempt_disable();
142                 if (tsk->thread.regs->msr & MSR_VEC) {
143 #ifdef CONFIG_SMP
144                         BUG_ON(tsk != current);
145 #endif
146                         giveup_altivec(tsk);
147                 }
148                 preempt_enable();
149         }
150 }
151
152 int dump_task_altivec(struct pt_regs *regs, elf_vrregset_t *vrregs)
153 {
154         flush_altivec_to_thread(current);
155         memcpy(vrregs, &current->thread.vr[0], sizeof(*vrregs));
156         return 1;
157 }
158 #endif /* CONFIG_ALTIVEC */
159
160 #ifdef CONFIG_SPE
161
162 void enable_kernel_spe(void)
163 {
164         WARN_ON(preemptible());
165
166 #ifdef CONFIG_SMP
167         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
168                 giveup_spe(current);
169         else
170                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
171 #else
172         giveup_spe(last_task_used_spe);
173 #endif /* __SMP __ */
174 }
175 EXPORT_SYMBOL(enable_kernel_spe);
176
177 void flush_spe_to_thread(struct task_struct *tsk)
178 {
179         if (tsk->thread.regs) {
180                 preempt_disable();
181                 if (tsk->thread.regs->msr & MSR_SPE) {
182 #ifdef CONFIG_SMP
183                         BUG_ON(tsk != current);
184 #endif
185                         giveup_spe(tsk);
186                 }
187                 preempt_enable();
188         }
189 }
190
191 int dump_spe(struct pt_regs *regs, elf_vrregset_t *evrregs)
192 {
193         flush_spe_to_thread(current);
194         /* We copy u32 evr[32] + u64 acc + u32 spefscr -> 35 */
195         memcpy(evrregs, &current->thread.evr[0], sizeof(u32) * 35);
196         return 1;
197 }
198 #endif /* CONFIG_SPE */
199
200 #ifndef CONFIG_SMP
201 /*
202  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
203  * and the current task has some state, discard it.
204  */
205 void discard_lazy_cpu_state(void)
206 {
207         preempt_disable();
208         if (last_task_used_math == current)
209                 last_task_used_math = NULL;
210 #ifdef CONFIG_ALTIVEC
211         if (last_task_used_altivec == current)
212                 last_task_used_altivec = NULL;
213 #endif /* CONFIG_ALTIVEC */
214 #ifdef CONFIG_SPE
215         if (last_task_used_spe == current)
216                 last_task_used_spe = NULL;
217 #endif
218         preempt_enable();
219 }
220 #endif /* CONFIG_SMP */
221
222 int set_dabr(unsigned long dabr)
223 {
224 #ifdef CONFIG_PPC_MERGE         /* XXX for now */
225         if (ppc_md.set_dabr)
226                 return ppc_md.set_dabr(dabr);
227 #endif
228
229         /* XXX should we have a CPU_FTR_HAS_DABR ? */
230 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
231         mtspr(SPRN_DABR, dabr);
232 #endif
233         return 0;
234 }
235
236 #ifdef CONFIG_PPC64
237 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
238 #endif
239
240 static DEFINE_PER_CPU(unsigned long, current_dabr);
241
242 struct task_struct *__switch_to(struct task_struct *prev,
243         struct task_struct *new)
244 {
245         struct thread_struct *new_thread, *old_thread;
246         unsigned long flags;
247         struct task_struct *last;
248
249 #ifdef CONFIG_SMP
250         /* avoid complexity of lazy save/restore of fpu
251          * by just saving it every time we switch out if
252          * this task used the fpu during the last quantum.
253          *
254          * If it tries to use the fpu again, it'll trap and
255          * reload its fp regs.  So we don't have to do a restore
256          * every switch, just a save.
257          *  -- Cort
258          */
259         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
260                 giveup_fpu(prev);
261 #ifdef CONFIG_ALTIVEC
262         /*
263          * If the previous thread used altivec in the last quantum
264          * (thus changing altivec regs) then save them.
265          * We used to check the VRSAVE register but not all apps
266          * set it, so we don't rely on it now (and in fact we need
267          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
268          *
269          * On SMP we always save/restore altivec regs just to avoid the
270          * complexity of changing processors.
271          *  -- Cort
272          */
273         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
274                 giveup_altivec(prev);
275 #endif /* CONFIG_ALTIVEC */
276 #ifdef CONFIG_SPE
277         /*
278          * If the previous thread used spe in the last quantum
279          * (thus changing spe regs) then save them.
280          *
281          * On SMP we always save/restore spe regs just to avoid the
282          * complexity of changing processors.
283          */
284         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
285                 giveup_spe(prev);
286 #endif /* CONFIG_SPE */
287
288 #else  /* CONFIG_SMP */
289 #ifdef CONFIG_ALTIVEC
290         /* Avoid the trap.  On smp this this never happens since
291          * we don't set last_task_used_altivec -- Cort
292          */
293         if (new->thread.regs && last_task_used_altivec == new)
294                 new->thread.regs->msr |= MSR_VEC;
295 #endif /* CONFIG_ALTIVEC */
296 #ifdef CONFIG_SPE
297         /* Avoid the trap.  On smp this this never happens since
298          * we don't set last_task_used_spe
299          */
300         if (new->thread.regs && last_task_used_spe == new)
301                 new->thread.regs->msr |= MSR_SPE;
302 #endif /* CONFIG_SPE */
303
304 #endif /* CONFIG_SMP */
305
306         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr)) {
307                 set_dabr(new->thread.dabr);
308                 __get_cpu_var(current_dabr) = new->thread.dabr;
309         }
310
311         new_thread = &new->thread;
312         old_thread = &current->thread;
313
314 #ifdef CONFIG_PPC64
315         /*
316          * Collect processor utilization data per process
317          */
318         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
319                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
320                 long unsigned start_tb, current_tb;
321                 start_tb = old_thread->start_tb;
322                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
323                 old_thread->accum_tb += (current_tb - start_tb);
324                 new_thread->start_tb = current_tb;
325         }
326 #endif
327
328         local_irq_save(flags);
329
330         account_system_vtime(current);
331         account_process_vtime(current);
332         calculate_steal_time();
333
334         last = _switch(old_thread, new_thread);
335
336         local_irq_restore(flags);
337
338         return last;
339 }
340
341 static int instructions_to_print = 16;
342
343 static void show_instructions(struct pt_regs *regs)
344 {
345         int i;
346         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
347                         sizeof(int));
348
349         printk("Instruction dump:");
350
351         for (i = 0; i < instructions_to_print; i++) {
352                 int instr;
353
354                 if (!(i % 8))
355                         printk("\n");
356
357 #if !defined(CONFIG_BOOKE)
358                 /* If executing with the IMMU off, adjust pc rather
359                  * than print XXXXXXXX.
360                  */
361                 if (!(regs->msr & MSR_IR))
362                         pc = (unsigned long)phys_to_virt(pc);
363 #endif
364
365                 /* We use __get_user here *only* to avoid an OOPS on a
366                  * bad address because the pc *should* only be a
367                  * kernel address.
368                  */
369                 if (!__kernel_text_address(pc) ||
370                      __get_user(instr, (unsigned int __user *)pc)) {
371                         printk("XXXXXXXX ");
372                 } else {
373                         if (regs->nip == pc)
374                                 printk("<%08x> ", instr);
375                         else
376                                 printk("%08x ", instr);
377                 }
378
379                 pc += sizeof(int);
380         }
381
382         printk("\n");
383 }
384
385 static struct regbit {
386         unsigned long bit;
387         const char *name;
388 } msr_bits[] = {
389         {MSR_EE,        "EE"},
390         {MSR_PR,        "PR"},
391         {MSR_FP,        "FP"},
392         {MSR_ME,        "ME"},
393         {MSR_IR,        "IR"},
394         {MSR_DR,        "DR"},
395         {0,             NULL}
396 };
397
398 static void printbits(unsigned long val, struct regbit *bits)
399 {
400         const char *sep = "";
401
402         printk("<");
403         for (; bits->bit; ++bits)
404                 if (val & bits->bit) {
405                         printk("%s%s", sep, bits->name);
406                         sep = ",";
407                 }
408         printk(">");
409 }
410
411 #ifdef CONFIG_PPC64
412 #define REG             "%016lx"
413 #define REGS_PER_LINE   4
414 #define LAST_VOLATILE   13
415 #else
416 #define REG             "%08lx"
417 #define REGS_PER_LINE   8
418 #define LAST_VOLATILE   12
419 #endif
420
421 void show_regs(struct pt_regs * regs)
422 {
423         int i, trap;
424
425         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
426                regs->nip, regs->link, regs->ctr);
427         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
428                regs, regs->trap, print_tainted(), init_utsname()->release);
429         printk("MSR: "REG" ", regs->msr);
430         printbits(regs->msr, msr_bits);
431         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
432         trap = TRAP(regs);
433         if (trap == 0x300 || trap == 0x600)
434 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
435                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
436 #else
437                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
438 #endif
439         printk("TASK = %p[%d] '%s' THREAD: %p",
440                current, current->pid, current->comm, task_thread_info(current));
441
442 #ifdef CONFIG_SMP
443         printk(" CPU: %d", smp_processor_id());
444 #endif /* CONFIG_SMP */
445
446         for (i = 0;  i < 32;  i++) {
447                 if ((i % REGS_PER_LINE) == 0)
448                         printk("\n" KERN_INFO "GPR%02d: ", i);
449                 printk(REG " ", regs->gpr[i]);
450                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
451                         break;
452         }
453         printk("\n");
454 #ifdef CONFIG_KALLSYMS
455         /*
456          * Lookup NIP late so we have the best change of getting the
457          * above info out without failing
458          */
459         printk("NIP ["REG"] ", regs->nip);
460         print_symbol("%s\n", regs->nip);
461         printk("LR ["REG"] ", regs->link);
462         print_symbol("%s\n", regs->link);
463 #endif
464         show_stack(current, (unsigned long *) regs->gpr[1]);
465         if (!user_mode(regs))
466                 show_instructions(regs);
467 }
468
469 void exit_thread(void)
470 {
471         discard_lazy_cpu_state();
472 }
473
474 void flush_thread(void)
475 {
476 #ifdef CONFIG_PPC64
477         struct thread_info *t = current_thread_info();
478
479         if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
480                 clear_ti_thread_flag(t, TIF_ABI_PENDING);
481                 if (test_ti_thread_flag(t, TIF_32BIT))
482                         clear_ti_thread_flag(t, TIF_32BIT);
483                 else
484                         set_ti_thread_flag(t, TIF_32BIT);
485         }
486 #endif
487
488         discard_lazy_cpu_state();
489
490         if (current->thread.dabr) {
491                 current->thread.dabr = 0;
492                 set_dabr(0);
493         }
494 }
495
496 void
497 release_thread(struct task_struct *t)
498 {
499 }
500
501 /*
502  * This gets called before we allocate a new thread and copy
503  * the current task into it.
504  */
505 void prepare_to_copy(struct task_struct *tsk)
506 {
507         flush_fp_to_thread(current);
508         flush_altivec_to_thread(current);
509         flush_spe_to_thread(current);
510 }
511
512 /*
513  * Copy a thread..
514  */
515 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
516                 unsigned long unused, struct task_struct *p,
517                 struct pt_regs *regs)
518 {
519         struct pt_regs *childregs, *kregs;
520         extern void ret_from_fork(void);
521         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
522
523         CHECK_FULL_REGS(regs);
524         /* Copy registers */
525         sp -= sizeof(struct pt_regs);
526         childregs = (struct pt_regs *) sp;
527         *childregs = *regs;
528         if ((childregs->msr & MSR_PR) == 0) {
529                 /* for kernel thread, set `current' and stackptr in new task */
530                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
531 #ifdef CONFIG_PPC32
532                 childregs->gpr[2] = (unsigned long) p;
533 #else
534                 clear_tsk_thread_flag(p, TIF_32BIT);
535 #endif
536                 p->thread.regs = NULL;  /* no user register state */
537         } else {
538                 childregs->gpr[1] = usp;
539                 p->thread.regs = childregs;
540                 if (clone_flags & CLONE_SETTLS) {
541 #ifdef CONFIG_PPC64
542                         if (!test_thread_flag(TIF_32BIT))
543                                 childregs->gpr[13] = childregs->gpr[6];
544                         else
545 #endif
546                                 childregs->gpr[2] = childregs->gpr[6];
547                 }
548         }
549         childregs->gpr[3] = 0;  /* Result from fork() */
550         sp -= STACK_FRAME_OVERHEAD;
551
552         /*
553          * The way this works is that at some point in the future
554          * some task will call _switch to switch to the new task.
555          * That will pop off the stack frame created below and start
556          * the new task running at ret_from_fork.  The new task will
557          * do some house keeping and then return from the fork or clone
558          * system call, using the stack frame created above.
559          */
560         sp -= sizeof(struct pt_regs);
561         kregs = (struct pt_regs *) sp;
562         sp -= STACK_FRAME_OVERHEAD;
563         p->thread.ksp = sp;
564
565 #ifdef CONFIG_PPC64
566         if (cpu_has_feature(CPU_FTR_SLB)) {
567                 unsigned long sp_vsid;
568                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
569
570                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
571                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
572                                 << SLB_VSID_SHIFT_1T;
573                 else
574                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
575                                 << SLB_VSID_SHIFT;
576                 sp_vsid |= SLB_VSID_KERNEL | llp;
577                 p->thread.ksp_vsid = sp_vsid;
578         }
579
580         /*
581          * The PPC64 ABI makes use of a TOC to contain function 
582          * pointers.  The function (ret_from_except) is actually a pointer
583          * to the TOC entry.  The first entry is a pointer to the actual
584          * function.
585          */
586         kregs->nip = *((unsigned long *)ret_from_fork);
587 #else
588         kregs->nip = (unsigned long)ret_from_fork;
589 #endif
590
591         return 0;
592 }
593
594 /*
595  * Set up a thread for executing a new program
596  */
597 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
598 {
599 #ifdef CONFIG_PPC64
600         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
601 #endif
602
603         set_fs(USER_DS);
604
605         /*
606          * If we exec out of a kernel thread then thread.regs will not be
607          * set.  Do it now.
608          */
609         if (!current->thread.regs) {
610                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
611                 current->thread.regs = regs - 1;
612         }
613
614         memset(regs->gpr, 0, sizeof(regs->gpr));
615         regs->ctr = 0;
616         regs->link = 0;
617         regs->xer = 0;
618         regs->ccr = 0;
619         regs->gpr[1] = sp;
620
621         /*
622          * We have just cleared all the nonvolatile GPRs, so make
623          * FULL_REGS(regs) return true.  This is necessary to allow
624          * ptrace to examine the thread immediately after exec.
625          */
626         regs->trap &= ~1UL;
627
628 #ifdef CONFIG_PPC32
629         regs->mq = 0;
630         regs->nip = start;
631         regs->msr = MSR_USER;
632 #else
633         if (!test_thread_flag(TIF_32BIT)) {
634                 unsigned long entry, toc;
635
636                 /* start is a relocated pointer to the function descriptor for
637                  * the elf _start routine.  The first entry in the function
638                  * descriptor is the entry address of _start and the second
639                  * entry is the TOC value we need to use.
640                  */
641                 __get_user(entry, (unsigned long __user *)start);
642                 __get_user(toc, (unsigned long __user *)start+1);
643
644                 /* Check whether the e_entry function descriptor entries
645                  * need to be relocated before we can use them.
646                  */
647                 if (load_addr != 0) {
648                         entry += load_addr;
649                         toc   += load_addr;
650                 }
651                 regs->nip = entry;
652                 regs->gpr[2] = toc;
653                 regs->msr = MSR_USER64;
654         } else {
655                 regs->nip = start;
656                 regs->gpr[2] = 0;
657                 regs->msr = MSR_USER32;
658         }
659 #endif
660
661         discard_lazy_cpu_state();
662         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
663         current->thread.fpscr.val = 0;
664 #ifdef CONFIG_ALTIVEC
665         memset(current->thread.vr, 0, sizeof(current->thread.vr));
666         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
667         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
668         current->thread.vrsave = 0;
669         current->thread.used_vr = 0;
670 #endif /* CONFIG_ALTIVEC */
671 #ifdef CONFIG_SPE
672         memset(current->thread.evr, 0, sizeof(current->thread.evr));
673         current->thread.acc = 0;
674         current->thread.spefscr = 0;
675         current->thread.used_spe = 0;
676 #endif /* CONFIG_SPE */
677 }
678
679 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
680                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
681
682 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
683 {
684         struct pt_regs *regs = tsk->thread.regs;
685
686         /* This is a bit hairy.  If we are an SPE enabled  processor
687          * (have embedded fp) we store the IEEE exception enable flags in
688          * fpexc_mode.  fpexc_mode is also used for setting FP exception
689          * mode (asyn, precise, disabled) for 'Classic' FP. */
690         if (val & PR_FP_EXC_SW_ENABLE) {
691 #ifdef CONFIG_SPE
692                 if (cpu_has_feature(CPU_FTR_SPE)) {
693                         tsk->thread.fpexc_mode = val &
694                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
695                         return 0;
696                 } else {
697                         return -EINVAL;
698                 }
699 #else
700                 return -EINVAL;
701 #endif
702         }
703
704         /* on a CONFIG_SPE this does not hurt us.  The bits that
705          * __pack_fe01 use do not overlap with bits used for
706          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
707          * on CONFIG_SPE implementations are reserved so writing to
708          * them does not change anything */
709         if (val > PR_FP_EXC_PRECISE)
710                 return -EINVAL;
711         tsk->thread.fpexc_mode = __pack_fe01(val);
712         if (regs != NULL && (regs->msr & MSR_FP) != 0)
713                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
714                         | tsk->thread.fpexc_mode;
715         return 0;
716 }
717
718 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
719 {
720         unsigned int val;
721
722         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
723 #ifdef CONFIG_SPE
724                 if (cpu_has_feature(CPU_FTR_SPE))
725                         val = tsk->thread.fpexc_mode;
726                 else
727                         return -EINVAL;
728 #else
729                 return -EINVAL;
730 #endif
731         else
732                 val = __unpack_fe01(tsk->thread.fpexc_mode);
733         return put_user(val, (unsigned int __user *) adr);
734 }
735
736 int set_endian(struct task_struct *tsk, unsigned int val)
737 {
738         struct pt_regs *regs = tsk->thread.regs;
739
740         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
741             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
742                 return -EINVAL;
743
744         if (regs == NULL)
745                 return -EINVAL;
746
747         if (val == PR_ENDIAN_BIG)
748                 regs->msr &= ~MSR_LE;
749         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
750                 regs->msr |= MSR_LE;
751         else
752                 return -EINVAL;
753
754         return 0;
755 }
756
757 int get_endian(struct task_struct *tsk, unsigned long adr)
758 {
759         struct pt_regs *regs = tsk->thread.regs;
760         unsigned int val;
761
762         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
763             !cpu_has_feature(CPU_FTR_REAL_LE))
764                 return -EINVAL;
765
766         if (regs == NULL)
767                 return -EINVAL;
768
769         if (regs->msr & MSR_LE) {
770                 if (cpu_has_feature(CPU_FTR_REAL_LE))
771                         val = PR_ENDIAN_LITTLE;
772                 else
773                         val = PR_ENDIAN_PPC_LITTLE;
774         } else
775                 val = PR_ENDIAN_BIG;
776
777         return put_user(val, (unsigned int __user *)adr);
778 }
779
780 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
781 {
782         tsk->thread.align_ctl = val;
783         return 0;
784 }
785
786 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
787 {
788         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
789 }
790
791 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
792
793 int sys_clone(unsigned long clone_flags, unsigned long usp,
794               int __user *parent_tidp, void __user *child_threadptr,
795               int __user *child_tidp, int p6,
796               struct pt_regs *regs)
797 {
798         CHECK_FULL_REGS(regs);
799         if (usp == 0)
800                 usp = regs->gpr[1];     /* stack pointer for child */
801 #ifdef CONFIG_PPC64
802         if (test_thread_flag(TIF_32BIT)) {
803                 parent_tidp = TRUNC_PTR(parent_tidp);
804                 child_tidp = TRUNC_PTR(child_tidp);
805         }
806 #endif
807         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
808 }
809
810 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
811              unsigned long p4, unsigned long p5, unsigned long p6,
812              struct pt_regs *regs)
813 {
814         CHECK_FULL_REGS(regs);
815         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
816 }
817
818 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
819               unsigned long p4, unsigned long p5, unsigned long p6,
820               struct pt_regs *regs)
821 {
822         CHECK_FULL_REGS(regs);
823         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
824                         regs, 0, NULL, NULL);
825 }
826
827 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
828                unsigned long a3, unsigned long a4, unsigned long a5,
829                struct pt_regs *regs)
830 {
831         int error;
832         char *filename;
833
834         filename = getname((char __user *) a0);
835         error = PTR_ERR(filename);
836         if (IS_ERR(filename))
837                 goto out;
838         flush_fp_to_thread(current);
839         flush_altivec_to_thread(current);
840         flush_spe_to_thread(current);
841         error = do_execve(filename, (char __user * __user *) a1,
842                           (char __user * __user *) a2, regs);
843         if (error == 0) {
844                 task_lock(current);
845                 current->ptrace &= ~PT_DTRACE;
846                 task_unlock(current);
847         }
848         putname(filename);
849 out:
850         return error;
851 }
852
853 #ifdef CONFIG_IRQSTACKS
854 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
855                                   unsigned long nbytes)
856 {
857         unsigned long stack_page;
858         unsigned long cpu = task_cpu(p);
859
860         /*
861          * Avoid crashing if the stack has overflowed and corrupted
862          * task_cpu(p), which is in the thread_info struct.
863          */
864         if (cpu < NR_CPUS && cpu_possible(cpu)) {
865                 stack_page = (unsigned long) hardirq_ctx[cpu];
866                 if (sp >= stack_page + sizeof(struct thread_struct)
867                     && sp <= stack_page + THREAD_SIZE - nbytes)
868                         return 1;
869
870                 stack_page = (unsigned long) softirq_ctx[cpu];
871                 if (sp >= stack_page + sizeof(struct thread_struct)
872                     && sp <= stack_page + THREAD_SIZE - nbytes)
873                         return 1;
874         }
875         return 0;
876 }
877
878 #else
879 #define valid_irq_stack(sp, p, nb)      0
880 #endif /* CONFIG_IRQSTACKS */
881
882 int validate_sp(unsigned long sp, struct task_struct *p,
883                        unsigned long nbytes)
884 {
885         unsigned long stack_page = (unsigned long)task_stack_page(p);
886
887         if (sp >= stack_page + sizeof(struct thread_struct)
888             && sp <= stack_page + THREAD_SIZE - nbytes)
889                 return 1;
890
891         return valid_irq_stack(sp, p, nbytes);
892 }
893
894 #ifdef CONFIG_PPC64
895 #define MIN_STACK_FRAME 112     /* same as STACK_FRAME_OVERHEAD, in fact */
896 #define FRAME_LR_SAVE   2
897 #define INT_FRAME_SIZE  (sizeof(struct pt_regs) + STACK_FRAME_OVERHEAD + 288)
898 #define REGS_MARKER     0x7265677368657265ul
899 #define FRAME_MARKER    12
900 #else
901 #define MIN_STACK_FRAME 16
902 #define FRAME_LR_SAVE   1
903 #define INT_FRAME_SIZE  (sizeof(struct pt_regs) + STACK_FRAME_OVERHEAD)
904 #define REGS_MARKER     0x72656773ul
905 #define FRAME_MARKER    2
906 #endif
907
908 EXPORT_SYMBOL(validate_sp);
909
910 unsigned long get_wchan(struct task_struct *p)
911 {
912         unsigned long ip, sp;
913         int count = 0;
914
915         if (!p || p == current || p->state == TASK_RUNNING)
916                 return 0;
917
918         sp = p->thread.ksp;
919         if (!validate_sp(sp, p, MIN_STACK_FRAME))
920                 return 0;
921
922         do {
923                 sp = *(unsigned long *)sp;
924                 if (!validate_sp(sp, p, MIN_STACK_FRAME))
925                         return 0;
926                 if (count > 0) {
927                         ip = ((unsigned long *)sp)[FRAME_LR_SAVE];
928                         if (!in_sched_functions(ip))
929                                 return ip;
930                 }
931         } while (count++ < 16);
932         return 0;
933 }
934
935 static int kstack_depth_to_print = 64;
936
937 void show_stack(struct task_struct *tsk, unsigned long *stack)
938 {
939         unsigned long sp, ip, lr, newsp;
940         int count = 0;
941         int firstframe = 1;
942
943         sp = (unsigned long) stack;
944         if (tsk == NULL)
945                 tsk = current;
946         if (sp == 0) {
947                 if (tsk == current)
948                         asm("mr %0,1" : "=r" (sp));
949                 else
950                         sp = tsk->thread.ksp;
951         }
952
953         lr = 0;
954         printk("Call Trace:\n");
955         do {
956                 if (!validate_sp(sp, tsk, MIN_STACK_FRAME))
957                         return;
958
959                 stack = (unsigned long *) sp;
960                 newsp = stack[0];
961                 ip = stack[FRAME_LR_SAVE];
962                 if (!firstframe || ip != lr) {
963                         printk("["REG"] ["REG"] ", sp, ip);
964                         print_symbol("%s", ip);
965                         if (firstframe)
966                                 printk(" (unreliable)");
967                         printk("\n");
968                 }
969                 firstframe = 0;
970
971                 /*
972                  * See if this is an exception frame.
973                  * We look for the "regshere" marker in the current frame.
974                  */
975                 if (validate_sp(sp, tsk, INT_FRAME_SIZE)
976                     && stack[FRAME_MARKER] == REGS_MARKER) {
977                         struct pt_regs *regs = (struct pt_regs *)
978                                 (sp + STACK_FRAME_OVERHEAD);
979                         printk("--- Exception: %lx", regs->trap);
980                         print_symbol(" at %s\n", regs->nip);
981                         lr = regs->link;
982                         print_symbol("    LR = %s\n", lr);
983                         firstframe = 1;
984                 }
985
986                 sp = newsp;
987         } while (count++ < kstack_depth_to_print);
988 }
989
990 void dump_stack(void)
991 {
992         show_stack(current, NULL);
993 }
994 EXPORT_SYMBOL(dump_stack);
995
996 #ifdef CONFIG_PPC64
997 void ppc64_runlatch_on(void)
998 {
999         unsigned long ctrl;
1000
1001         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1002                 HMT_medium();
1003
1004                 ctrl = mfspr(SPRN_CTRLF);
1005                 ctrl |= CTRL_RUNLATCH;
1006                 mtspr(SPRN_CTRLT, ctrl);
1007
1008                 set_thread_flag(TIF_RUNLATCH);
1009         }
1010 }
1011
1012 void ppc64_runlatch_off(void)
1013 {
1014         unsigned long ctrl;
1015
1016         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1017                 HMT_medium();
1018
1019                 clear_thread_flag(TIF_RUNLATCH);
1020
1021                 ctrl = mfspr(SPRN_CTRLF);
1022                 ctrl &= ~CTRL_RUNLATCH;
1023                 mtspr(SPRN_CTRLT, ctrl);
1024         }
1025 }
1026 #endif